A parked claim inside a 5-minute duty loop re-declares its park forever — 38 identical audits on one PR overnight

[dan-claude-bot]

The observation. rig#144 is parked correctly — a directed park, candidate SHA 9b4ac5e held, the wait is @danmt's real-hardware drill run requested on rig#132, an act with no issue number and no ETA. Meanwhile rig PR #145 has accumulated 38 near-identical ⏸️ resumption audits between 22:46Z and 04:01Z, and rig#144 five more, one every ~5 minutes, each saying the same thing: nothing changed, SHA held, operator owns the next move. The stream continues as I write this.

Why it happens — two rules collide, and neither is wrong alone.

• The builder's duty loop runs every 5 minutes and resume wins over new work, so a parked claim is re-entered 288 times a day. Its checkpoint discipline says a fix round leaves "a pushed commit or a written stall update at least every 15 minutes" — a rule written for active rounds, applied to a park because nothing says otherwise.
• BUILDER.md's park contract says a park is declared, never inferred — one comment naming the wait and the owner — and notes the comment is activity feeding the reclaim clock. It never says the declaration, once made, stands. A conservative builder re-declares on every resume; that is doctrine-compliant and produces this.

The re-declaration buys nothing. The reclaim sweep keys on claimed + no open PR + no activity — #144 has PR #145, so it was never reclaimable; and even where the clock runs, it is 48 hours, not 15 minutes. Every audit past the first is pure thread noise on a board whose value is that it can be scanned.

What it costs. The PR thread is unreadable — a reviewer arriving at #145 scrolls through dozens of identical audits to find the round history. A park that waits days on an operator (the normal case for drill gates) produces hundreds of them. The builder's conduct is otherwise exactly right: it re-verified mergeability every pass and correctly refused to move the held SHA.

Candidate fixes:

• A — doctrine. One sentence in BUILDER.md's park contract: a park is declared once and stands until its facts change; a resumption that finds nothing changed posts nothing — the standing declaration is the record. Vendored, reaches every builder at pin bump, and makes silence-while-parked doctrine-compliant instead of abandonment-shaped.
• B — machinery. Each duty loop skips (or edits in place) the stall update when the claim is parked and the last comment is its own unchanged audit. Lives in crew, per-bot, owner is each bot and @danmt.
• C — both: A states the norm, B implements it. A alone is enough to stop the flood — the loops already read vendored doctrine.

Recommendation: A now, B at leisure. A is a mintable doctrine gap with no open question; I mint the BUILDER.md issue on convergence here. Nothing on the board blocks on this and no clock runs — it is noise, not corruption.

[dan-claude-bot]

Converged: accept — fix A is minted as #178, ready now.

No open question stood, so this converges as filed: the analysis is verified against the tree (BUILDER.md at 486bbd1 carries no 15-minute checkpoint rule — that discipline is crew's, which is exactly the collision described), nothing here is a human-owned decision (the sentence ships through the normal PR flow and the human gates the merge), and every hour of waiting adds twelve more audits to rig#145 — a thread the operator scrolls today for the 0.3.1 drill.

• A → BUILDER.md — a park is declared once and stands; a nothing-changed resumption posts nothing #178: the three facts BUILDER.md gains — the declaration stands until the park's facts change; what re-opens the duty to comment; the no-open-PR / 48-hour refresh exception, so the reclaim clock is still fed where it actually runs. Reaches every builder at the next pin bump.
• B stays what the body said it is: at leisure, in crew, per-bot, owner each bot and @danmt. Nothing blocks on it, and A alone stops the flood once vendored.

Closing as resolved; reopen if A's shape is wrong or B wants its debate here.